LiDAR-based 3D object detection achieves superior performance. However, the unevenly distributed point clouds on foreground objects can weaken their geometric representation. Meanwhile, far-away objects typically have very few points, which further impairs detection performance. In this article, a novel framework PUDet is presented, which integrates generative models into discriminative detectors. A point cloud upsampling network is leveraged with prior knowledge to enhance the geometric details of foreground objects, aiding the detector in achieving more accurate prediction. PUDet incorporates two key modules: LDEM for nearby objects, which optimizes point distribution while minimizing computational costs, and DDAM for distant objects, which increases point density to better delineate object contours. To evaluate the optimization of geometric contours, the uniform loss of close and long-distance targets before and after enhancement is experimentally compared, showing the efficacy of LDEM and DDAM. This article also displays the attention maps on object point clouds, explaining the observed accuracy gains. Experimental results on the KITTI testing set show that the proposed framework improves the baseline CT3D by 1.84 mAP, confirming the effectiveness of PUDet. This work introduces a novel approach to 3D object detection, enhancing precision and reliability in object recognition for applications like autonomous driving.